A collection of tools for analyzing data from my eyeCode experiment
Python
Fetching latest commit…
Cannot retrieve the latest commit at this time.
Permalink
Failed to load latest commit information.
data
img
programs
Makefile
README.md
grade_trial.py
notebook2markdown.py
pretty_plot.py
response.ipynb
response.md
response_stats.py
tobii_count_valid.py
tobii_filter_valid.py
tsv2csv.py

README.md

eyeCode Tools

A collection of tools for analyzing data from the eyeCode experiment. Please cite the linked paper if you use this data set!

Author: Michael Hansen (mihansen@indiana.edu)

Scripts

There are two main scripts for the data analysis:

  1. grade_trial.py - Attempts to automatically grade a single trial (see code for a list of common errors)
  2. response_stats.py - Plots results and computes some simple stats

Data Format

The eyeCode response data set is available in the data directory as an XML file. The format of the response data XML is described below.

  • <experiments> - top-level element
    • created - date and time when the file was created (format = %Y-%m-%d %H:%M:%S)
    • <experiment> - a single experiment (10 trials)
      • started - date and time when experiment started (including surveys, format = %Y-%m-%d %H:%M:%S)
      • ended - date and time when experiment ended (including surveys, format = %Y-%m-%d %H:%M:%S)
      • <questions> - all survey questions
        • <question> - a single survey question
          • name - name of question, one of:
            • age - Participant's age in years
            • correct - How many programs did you correctly predict? Response is one of "few", "half", "most".
            • difficulty - How hard were the programs? Response is one of "easy", "medium", "hard".
            • education - What is the highest degree you've received? Response is one of "none", "bachelors", "masters", "phd", "other".
            • gender - Participant's gender. Response is one of "male", "female", "unreported"
            • major - Are you (or were you) a Computer Science major? Response is one of "no", "current", "past".
            • programming_years - Years of overall programming experience (may be fractional)
            • python_years - Years of Python experience (may be fractional)
          • value - participant's answer
      • <trials> - all trials (should be 10, unless some were discarded)
        • <trial> - single trial with one program
          • started - date and time when trial started (format = %Y-%m-%d %H:%M:%S)
          • ended - date and time when trial ended (format = %Y-%m-%d %H:%M:%S)
          • language - should always be python
          • order - numerical ordering of the trial (starting at 0)
          • base - program base type (there were 10 total base types)
          • version - version of the program (2-3 per base)
          • grade-value - response quality from 0-10
          • grade-category - category of grade, one of:
            • correct exact - response matched correct output, character-for-character
            • correct lines - response formatting was off, but had the right values and number of lines
            • correct values - response values matched, but not formatting or lines
            • common \* - response was incorrect, but matched a common error (exact/lines/values same as correct)
            • manual - response was graded manually
          • response-duration - number of seconds between first and last response keystrokes
        • <metrics> - code metrics for this trial's program
          • <metric> - single code metric
            • name - name of metric. One of:
              • code chars - number of characters in the program
              • code lines - number of lines in the program
              • cyclomatic complexity - McCabe's CC metric
              • halstead effort - E from Halstead software metrics
              • halstead volume - V from Halstead software metrics
              • output chars - number of characters in the true output
              • output lines - number of lines in the true output
        • <true-output> - what the program actually outputs
        • <predicted-output> - the participant's predicted output